Lyra

A macOS menu-bar app that shows synced, translated lyrics for whatever you're playing in Spotify. Local-first, private, and built to stay out of your way.

Lyra watches the Spotify desktop app, fetches time-synced lyrics, and translates them to English on-device using a local LLM, then floats them in a frosted panel that sits above your other windows. Everything happens on your machine: no accounts, no API keys, no cloud.

This is a personal project, shared as source. It is not distributed as a downloadable app (see Building).

Features

• Synced lyrics that follow the current Spotify track, line by line.
• On-device English translation via a local Ollama model. Fully-English lines are left as-is; non-English and mixed lines are translated.
• Two themes: a neutral frosted glass look, and an immersive theme that uses the current song's album art, blurred, as a living backing that changes with each track.
• A floating panel you can drag anywhere and resize, with its position, size, and theme remembered between launches.
• Album art in the header, and a glanceable "Title · Artist" in the menu bar when the panel is hidden.
• Offline caching: once a song is translated, it loads instantly on every replay.
• Graceful degradation: if the translation model isn't running, lyrics still show in their original language.

How it works

Lyra reads the currently playing track from the Spotify desktop app over AppleScript (no Spotify account or web API involved). It fetches time-synced lyrics from LRCLIB, with a strict matching gate so it never shows the wrong song's lyrics. Each song is translated in a single pass by a local Ollama model, with results cached in a local database so repeats are instant.

Nothing leaves your machine except the lyric lookup to LRCLIB and the album-art fetch. There are no API keys, no analytics, and no cloud translation.

Requirements

• macOS (Apple Silicon or Intel).
• The Spotify desktop app.
• Ollama running locally with a translation-capable model (developed against qwen2.5:7b; a larger model improves quality, especially for romanized lyrics).

Building

Lyra is shared as source and is not distributed as a prebuilt app. To run it, build it yourself:

# install dependencies
npm install
 
# run in development
npm run tauri dev
 
# build a local .app bundle
npm run tauri build

On first launch macOS will ask for permission to control Spotify (for reading the current track). An unsigned local build will also prompt Gatekeeper once; right-click the app and choose Open to get past it.

Technical details

• Built with Tauri v2: a React + TypeScript + Vite frontend with a thin Rust core that owns the window, tray, and OS integration.
• Spotify state is read via AppleScript from Rust and polled once a second, with sub-second interpolation for smooth lyric progress.
• Synced lyrics come from LRCLIB via an exact lookup, then a strict-gated search (normalized title plus duration tolerance) so a near-miss never resolves to the wrong song.
• Translation uses a local Ollama model in a single whole-song call, with a numbered transport so individual lines stay aligned and a per-line fallback so a dropped line never desyncs the rest.
• Lyrics and translations are cached in SQLite; only validated results are stored.
• The frosted panel uses native macOS vibrancy (not CSS), with album art delivered through the Rust layer so the app's network surface stays scoped to LRCLIB and the local model only.

Acknowledgements

• LRCLIB by tranxuanthang, the open lyric library Lyra depends on for synced lyrics.
• Ollama for making local model inference easy.
• Tauri and the window-vibrancy crate.